Building AI-Powered Software for Hospitals

Let's start with the first thing.

Introducing Heasy

My name is Remy. I'm the AI founding engineer at Heasy.

Before, Juliette presented us as a stealth startup because it's been only eight months. However, in these eight months, we have already built so much and we've already raised 2 million euros all in our pre -seed round.

In Heasy, our main goal is to build a software from the ground up for healthcare and hospitals. And now we are already working with the leading healthcare facilities all over Europe.

We're a team of 14 people and we are still growing. Some of us are in Spain, some others are in Italy, and we always travel into different countries to meet with healthcare experts, which are also part of our team.

A Faster, More Efficient Hospital Platform

Our main goal is to provide hospitals and clinics with a software that is faster, more easy to use, and more efficient. And one of the ways we do this is using AI.

There's so many ways to use AI, I feel like, today. day, and the main one we are focusing about, is generative UI. I will show you some examples in just a little bit.

And after those examples, which will help you understand how this helps the operators navigate the very complicated tasks they have to use in a more visual way, I will also get into privacy and safety, because AI is so interesting and so helpful.

However, we're dealing with clinical data from patients, and we cannot let this data go all around the world, to different countries and like this.

Generative UI for Healthcare Operations

So this is just an overview of our platform, a pretty standard platform with fixed pages that can help the operators do actions on the patient, on the waiting list, on the wards and the beds, and how they work.

However, the most interesting part is our AI assistant.

Moving Beyond Traditional Chatbots

For ease of use today, I've added a couple of prompts already here. I so I don't have to type everything.

And we are all used to the typical AI chatbots where you write some text, and it answers some texts. And this is OK when studying and researching.

But when you have to do complicated actions, and you have to move people from a person from a room to another room, or when you have to delete an appointment, having a graphic result, I think it makes everything more easy.

For example, if I want to find a patient, I can just see the name, the surname, and also maybe his date of birth. However, with our generative UI, a widget will be generated in real time with the data of that person.

Nothing is hard -coded or fixed like it was before, but it gets generated and it also offers the possibility to interact with it. For example, here we have this patient, the data, and we already, once we find it, we already have the option to prepare the transfer of this person.

And again, we are not hard -coded and limited to a typical fixed page, but this is going to be generated with the use of AI and everything in real time, as you can see. And in this case, we can also have a confirmation of what's happening and where she's being moved.

Real-Time Operational Insights

Some other examples are which words are the busiest. Instead of going from page to page, from section to section, We can use AI to get all of the data and see it visually represented in real time in our screen.

Another example can be see today's discharges. So instead of having to filter in the page and looking at the result, we have a clear view of what's happening. And we can already discharge someone who's ready. Carlo Rossi, for example, he's the first one at 11 a .m. So I'm already suggested to do this.

and here we will generate another component that will help us do this easier. We always have an overview of what's happening and it's always generated and updated in real time.

Privacy-First AI Architecture

I will not bore you too much with showing you all of the different features and that it actually can answer you and everything but I would like to dive more in particular on the privacy and the architecture.

It's not gonna be too complicated, this is is definitely too complicated.

What you will like more is this example here.

Protecting Clinical Data Through Anonymization

When we launch a request to our AI, for example, admits this particular person, Sophia, to this particular organization, orthopedics, on this data, we don't want to give this data to an LLM provider who's outside. So we've created this system of anonymization.

Let me show you what will happen. if this is the request written by the operator in the hospital we will have an anonymization which we run on our local model which is completely safe and controlled by us before sending the message and the request outside and so

our message will become like this admit this person to a ward on a particular date and we will use the AI only to classify our kind of request and the The answer will be generated by our code in the old style, so all the data is private and is safe.

Keeping Patient Data Local

Our question will be classified, as I told you before, so admissions create. We will create a particular payload that is kept locally. We will retrieve the personal data of this person,

so we only got the admission function from the AI, and then we have the personal data always kept inside of our software and after this the code works and we get this piece of

Improving Hospital Efficiency and Care

code always generated we are trying to work and keep implementing this and finding new ways to interact with the software and generating it in real time

we think this will make it much easier for our clients and for all the hospital to work better and more efficient way and to save time that now is wasted and and save money and also save more lives at the same time.

I hope I wasn't too fast or too long. This was everything. And thank you so much for listening to me.